Musical Discovery

For much of the latter half of the 20th century, new music discovery went something like this. An artist would make a song and they’d send demo tapes out to record companies and radio stations. They’d play to dimly lit bars and clubs, hoping that an A&R impresario lurked in the crowd. If they were lucky, a DJ might listen to their demo and would play it live. Perhaps someone would record it and start bootlegging tapes. These contraband tapes would be passed around and listened to by teenagers gathered in bedrooms. If all went well, the artist’s popularity would grow. They’d be signed, played more on the radio and do bigger shows. Fans and soon-to-be fans would go to record stores to listen to the new releases and buy the music on vinyl, tape or CD. The record shops would make money, the musicians would make money, the record companies would make more money.

This began to shift in the early naughties, driven, as so much was, by the emergence of the internet. The newfound ability to rip CDs and transform tracks into easily shareable mp3s on the likes of Napster rendered the entire world’s music repertoire available gratis to eager ears. For those that preferred their music to come without the lawbreaking, the iTunes store and others made purchasing it just about as easy. MP3 players and the iPod made it effortless to carry 1000 songs in your pocket. The days of physically owning your music were all but over in the space of only a few short years.

Despite the music industry’s hope that killing Napster would stem the rising tide, the death of the platform only resulted in more alternatives appearing. It turned out that people liked having instant access to all music for pretty much free. Music discovery underwent a transformation. To acquire a song, one simply had to search for it, and within minutes, it was yours—provided Kazaa or uTorrent were operational and your parents didn’t pick up the phone and break the connection. Online forums teemed with enthusiasts discussing new musical revelations and leaks, offering nearly everything and anything you desired, all for free.

Music was no longer scarce; there were effectively infinite copies of every single song in the world to which anyone could have immediate access. Gone were the days of friends passing around tapes or lingering in record stores. The social aspect of music discovery shifted from smoky bars, intimate bedrooms, and record emporiums to the virtual amphitheaters of online forums, Facebook threads, and text message exchanges.

The big problem with all of this of course was that it was all quite illegal.

In 2006, Daniel Ek was working at uTorrent. He could see that users wouldn’t stop pirating music and that the attempts by the Music Industry to thwart sharing were doomed to failure. He “realized that you can never legislate away from piracy. Laws can definitely help, but it doesn’t take away the problem. The only way to solve the problem was to create a service that was better than piracy and at the same time compensates the music industry – that gave us Spotify.”

Musical Curation

Spotify launched with the simple headline: A world of music. Instant, simple and free.

By 2023 it has over 500 million users.

For many music fans, playlists took center stage, with enthusiasts investing hours trawling the internet for them. Spotify introduced a feature to see what your friends were listening to via Facebook and then to directly share playlists with others. Then they made playlists searchable. That killed off sites like sharemyplaylist, but meant that when I needed three hours of Persian Wedding songs, all I had to do was hit the search bar and appear to be intimately familiar with the works of Hayedeh and Dariush for my then soon-to-be in-laws.

In 2015 Spotify launched Discover, a dynamic playlist which introduced users to tracks that were similar to what the listener had played recently. It was remarkably good. The social aspect of music discovery was being lost but it was replaced with an automaton that did the job exceptionally well, even if the results were sometimes corrupted by the false signal of repeated plays of Baby Shark.

What was more subtle about what had been happening throughout this period was that the way people consumed music was changing. We had progressed from music discovery as a purposeful act to one in which it was an every day occurrence. Background music had always existed, but it was via the radio, or compilations. This was personal. The value of the music itself transformed. The ability to have a consistent soundtrack playing, at home, at work, in the car or as you made your way through every day life, meant that listeners weren’t necessarily concerned about the specific artists that were playing, they had become more interested in the general ambience of that ever-present background music. Listeners still certainly relished the release of the new Taylor Swift album, but they also listened to music that they didn’t know more easily and without ever inquiring as to who the artist was, simply because it fit within the soundtrack of their lives.

The Discover feature was one of Spotify’s first public forays into personalised music curation using machine learning. The success of the project led to more experiments. It turned out that people loved the feature.

Spotify in 2023 is remarkable. When I want to run to rock music, the tempo and enthusiasm of the suggested playlist is exactly right. When I want to focus on work, the beats are mellow and unobtrusive. The playlists “picked for me” change daily, powered by AI. I still create my own playlists, but the experience is now akin to using ChatGPT. I add a few songs to set a general mood and Spotify offers up suggestions that match the general vibe. Prior to a recent trip, I created a playlist called “Italy Background Music”, which Spotify duly filled with tracks I wouldn’t have had the first idea about where to find. They were exactly what I was looking for.

Curation and general discovery, it seems, have been broadly solved by Spotify.

Musical Creation

I’ve become accustomed to hearing tracks that I’ve never before heard and wouldn’t have the first idea about the artists of. Occasionally, I’ve tapped through on an unknown song and discovered that it has only had a couple of hundred thousand plays, ever. Spotify is clearly drawing on the entire breadth of artists within its library to match my musical preferences. Or is it?

In 2016, Music Business Worldwide published an article stating:

Spotify is starting to make its own records.

Multiple cast-iron sources have informed us that, in recent months, Daniel Ek‘s company has been paying producers to create tracks within specific musical guidelines.

By introducing its own music into the (literal) mix, for which it has paid a producer a flat fee, has added to the platform under a false name and then surfaced to listeners via its AI curated playlists, the platform is solving two issues that it considers important: from a user’s perspective, more music that fits their desired soundscape is a good thing. From Spotify’s perspective, having the ability to add, say, a 3 minute track of in-house music (which they don’t need to pay royalties on) to every hour of listening means that their cost for that hour is reduced by 5%. The losers in this case are the artists, who would otherwise have earned from that three minute play.

There is nothing that says that Spotify can’t do this though, it’s their platform and even when big artists have removed their music from the platform in protest at the company’s policies or actions, they’ve quietly reappeared within months, such is the value of the service¹.

In the above 2016 article, it is clear that the firm was paying actual producers to make the music. In 2023, the landscape is likely quite different. AI has advanced to such a point where the beats, melodies and riffs of jazz, trip hop and other non-vocal music can be quite easily produced by a well trained AI. There are dozens of sites that algorithmically generate lo-fi background music. If Spotify isn’t already adding tracks generated by AI that perfectly match a given vibe, especially within those non-vocal musical genres, then it is at least experimenting with it. The prize is too large to not. In 2021, the company paid more than $7bn to rights holders. At 5%, that’s a nice $350m to find down the back of the AI sofa.

Licensing

Where this leads in my mind is something sort of entirely new.

Whilst vocal-less (can we really call it instrumental?) music is the easiest use of the technology, we saw earlier this year as there was a brief explosion of AI generated tracks from creators using AI voice models that imitated the likes of Drake and Kanye. Whilst these tracks weren’t perfect, they showed an early preview of technology that will change the face of music. The Hugging Face community is full of models of popular artists which can replicate the sound of a given singer or rapper and it is evident improvements move at a rapid clip, with some now indistinguishable from the original artist.

Licensing of brands exists broadly in most other industries. In fashion it saved Ralph Lauren (although nearly killed Burberry). It famously turned Nike from pure sportswear to casual fashion mega-brand. Could we see the emergence of the artist as a brand? The potential for artists to either directly license their musical likeness to a given platform or to allow producers to use an authorised AI model of their voice to create tracks which they, or their team, would have final sign off on could allow vocalists to extend their reach drastically.

Whilst the last idea might sound fanciful, there are artists who already draw on the online community. One of the DJ/Producer Deadmau5’s biggest tracks was the result of a fan sending him a demo of a vocal mix via twitter for a track that Deadmau5 had produced the previous day in a livestream.

We’ve also seen a rash of artists selling their music rights–will the future see those artists who reach the end of their careers sell their “official” AI model to allow them and their families to earn in perpetuity? It’s been proven repeatedly that those artists who adapt to the changing world are the ones that succeed, but this is something entirely new.

What seems certain however is that the music that we listen to in the coming years will be picked for us by machines and at least partially created using AI.

Footnotes

1. There is a good argument that Taylor’s reasoning for removing her music wasn’t entirely to do with this. ↩